Generated Model Based Data Augmentations for Classification of HER2 Immunohistochemical Pathological Images in Breast Cancer
Yeting Ma, Fangbo Yu, Peiqin Feng, Yali Zheng, Zhengnan Wang · 2021
In recent years, deep learning technology is widely used in the study of pathological diagnosis. Deep learning technology has a very positive impact on pathology. It has a strong ability to learn pathological images and can solve complex problems. But in order for models using deep learning technology to achieve better performance, we need a large number of datasets to train. However, it is difficult for the model to obtain labeled pathological image training samples, which leads to a small dataset. This will lead to the occurrence of overfitting problem, which will reduce the generalization ability of the model [1].In this paper, we show that utilizing different data augmentation techniques can alleviate these problems. By comparing the effects of different data augmentation techniques on the classification results of HER2 immunohistochemical(IHC) pathological images dataset in breast cancer, we show that the use of data augmentations techniques can help improve the classification results of pathological images. In particular, we found that the generation model was most effective for HER2 IHC pathological image classification, where DCGAN showed stronger performance.